Xie Qihang, Qu Haoran, Li Jianfeng, Zeng Rui, Li Wenhao, Ouyang Rui, Zhang Chengxiang, Xie Siyu, Du Ming
Department of Cardiothoracic Surgery, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Department of General Medicine, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Front Public Health. 2025 Jun 25;13:1600729. doi: 10.3389/fpubh.2025.1600729. eCollection 2025.
Emphysema is a major contributor to lung disease progression and is associated with significant health risks, including exacerbations, mortality, and lung cancer. While environmental exposures, such as brominated flame retardants (BFRs), have been suggested as risk factors, their role in emphysema prediction has been largely overlooked. This study aimed to develop a machine learning (ML) model to predict emphysema risk incorporating BFRs exposure data and demographic characteristics.
Using data from the NHANES (2005-2016) dataset, 8,205 participants were included in the study. The participants were divided into a training set (70%) and a testing set (30%). Eight machine learning algorithms, including lightGBM, MLP, DT, KNN, RF, SVM, Enet, and XGBoost, were applied to build and evaluate the model. Demographic data and BFRs exposure levels were used as predictors. SHAP and Partial Dependence Plots (PDP) were used for model interpretability analysis.
The MLP model showed the best performance with an AUC of 0.83. Age and PBB153 were identified as the most influential predictors. SHAP analysis revealed that higher exposure to BFRs, particularly PBB153, was strongly associated with increased emphysema risk. The WQS model further confirmed the positive relationship between BFRs exposure and emphysema.
This study demonstrates the significant predictive value of BFR exposure in emphysema risk assessment and highlights the importance of incorporating environmental factors into disease prediction models. The findings provide new insights for integrating BFRs into personalized health risk assessments and public health interventions.
肺气肿是导致肺部疾病进展的主要因素，与包括病情加重、死亡率和肺癌在内的重大健康风险相关。虽然环境暴露因素，如溴化阻燃剂（BFRs），已被认为是风险因素，但其在肺气肿预测中的作用在很大程度上被忽视了。本研究旨在开发一种机器学习（ML）模型，以结合BFRs暴露数据和人口统计学特征来预测肺气肿风险。
利用美国国家健康与营养检查调查（NHANES，2005 - 2016年）数据集的数据，8205名参与者被纳入本研究。参与者被分为训练集（70%）和测试集（30%）。应用包括轻量级梯度提升机（lightGBM）、多层感知器（MLP）、决策树（DT）、k近邻算法（KNN）、随机森林（RF）、支持向量机（SVM）、弹性网络（Enet）和极端梯度提升（XGBoost）在内的八种机器学习算法来构建和评估模型。人口统计学数据和BFRs暴露水平被用作预测因子。使用SHAP值和局部依赖图（PDP）进行模型可解释性分析。
MLP模型表现最佳，曲线下面积（AUC）为0.83。年龄和2,2',4,4',5,5'-六溴联苯（PBB153）被确定为最具影响力的预测因子。SHAP分析表明，较高的BFRs暴露，尤其是PBB153，与肺气肿风险增加密切相关。加权分位数和（WQS）模型进一步证实了BFRs暴露与肺气肿之间的正相关关系。
本研究证明了BFRs暴露在肺气肿风险评估中的显著预测价值，并强调了将环境因素纳入疾病预测模型的重要性。这些发现为将BFRs纳入个性化健康风险评估和公共卫生干预提供了新的见解。
